Update MVDR beamforming tutorial #2398
Conversation
|
@nateanl has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
examples/tutorials/mvdr_tutorial.py
Outdated
| @@ -1,32 +1,45 @@ | |||
| """ | |||
| MVDR with torchaudio | |||
| ==================== | |||
| MVDR Beamforming with TorchAudio | |||
There was a problem hiding this comment.
I have been feeling with TorchAudio is redundant.
Can we rephrase it to <Task> with MVDR Beamforming?
From the content of the tutorial, I think Noise reduction with MVDR Beamforming is a good title.
examples/tutorials/mvdr_tutorial.py
Outdated
| waveform_clean, sr2 = torchaudio.load(SAMPLE_CLEAN) | ||
| assert sr == sr2 == SAMPLE_RATE | ||
| # The mixture waveform is a combination of clean and noise waveforms | ||
| waveform_noise = waveform_mix - waveform_clean |
There was a problem hiding this comment.
This part felt unintuitive to me.
If it were waveform_mix = waveform_clean + waveform_noise then it makes more sense.
Can we create another asset and fetch it in the same way as the rest?
examples/tutorials/mvdr_tutorial.py
Outdated
| magnitude = stft.abs() | ||
| spectrogram = 20 * torch.log10(magnitude + 1e-8).numpy() | ||
| figure, axis = plt.subplots(1, 1) | ||
| img = axis.imshow(spectrogram, cmap="nipy_spectral", vmin=-100, vmax=0, origin="lower", aspect="auto") |
There was a problem hiding this comment.
The custom colormap is cool, but can we use the default viridis?
It is designed so that it is easy for people with color blindness to understand the plot.
See the story behind the development https://bids.github.io/colormap/
|
@nateanl has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
Co-authored-by: Caroline Chen <carolinechen@fb.com>
Co-authored-by: Caroline Chen <carolinechen@fb.com>
|
@nateanl has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
Hey @nateanl. |
download_assetto download audios.MVDRmodule with new-addedSoudenMVDRandRTFMVDRmodules.F.rtf_evdandF.rtf_powerfor RTF computation.